This ranks and warns. It does not tell you what to buy.
Tested on 234 trading days the model had never seen (Sep 2025 → Aug 2026): the AVOID flag worked — flagged stocks returned −1.68% against −0.50% for the rest. The 60-day ranking on tradeable stocks did not hold up (correlation −0.001). Short-horizon ranking held but is too weak to cover the ~2.2% round-trip cost of trading.
In every test, buying a spread of stocks and holding beat every active strategy. Treat the ranking as a starting point for your own research.
| Symbol | Name | Conv | 1d | 5d | 20d ▼ | 60d | Unusual | Liquidity | 20d ret | From high | News | Flags |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||||||
1d / 5d / 20d / 60d are percentiles within the day (1.00 = highest rated), each from the model trained on that horizon. The buttons above choose which one the Buy and Sell lists are cut from — top and bottom decile respectively. Conv is agreement between differently-seeded models — low means the ordering is close to arbitrary. Unusual is 0–100 describing how the stock traded against its own history: volume spikes, price ramps, moves on a thin book. AVOID is a separate, older flag: the bottom decile of a weighted blend of all four horizons. It is kept because it is the one signal measured on unseen data, and it is the flag the −1.68% figure above refers to — the per-horizon Sell lists have not been validated separately.
The unusual score is descriptive only. It measures trading patterns. It identifies no person, asserts no intent, and is not evidence of wrongdoing — the same footprint precedes genuine accumulation and a pump, which is why it is shown beside the ranking rather than blended into it.